Research On Free Virtual View Generation For3DTV Systems

Compared with traditional two-dimensional television(2DTV),three-dimensional television(3DTV) expands the viewers’ experience about the real stereoscopic world and achieves interaction between users and television scene due to its scene depth information.3DTV has been a new developing direction after high-definition plate television.The multi-view stereoscopic feeling depends on the several video image streams and virtual view generation is one of the most important technology to realize it.Depth image-based rendering(DIBR) is a process that uses one2D color image and its associated depth image to render virtual view3D images.Firstly,this paper introduces the main technology of3DTV systems.Then we focuses on the research of virtual free-view generation for3DTV systems.In order to generate high quality virtual view images,the paper utilizes the technology of DIBR,which is considered having much potential for view rendering.In this paper we propose three new methods about how to generate virtual views.A new method named Euclidean-distance-and-depth Based the Pixel Allocation(EBPA) is proposed to solve the problems of overlapping and resampling occured during DIBR.This method can successfully remove the tiny holes by comparing the depth of non-integer pixel around tiny holes in certain a range and Euclidean-distance. Experiments show that EPBA can effectively eliminate the small holes and highly improve the quality of the virtual viewimage.In this paper, an eight-direction straight line detection method to fill holes for a simple background is proposed.This method completes the line detection of eight directions by the pixels information around the holes,then predicting the hole region and completing the hole filling. Compared to the methods of general level hole filling and Gaussian Filtering on the depth map, the experimental simulation shows that this algorithm improves the image quality both on subjective feelings and objective parameters.In real life, the background texture is generally more complex, but the depth of information is relatively simple.In view of this situation, this paper proposes a depth image-based inverse rendering to fill the holes. We can get a high-quality depth image after simple level hole filling and then projects the hole pixels onto the image plane of the auxiliary reference point of view.Compare with the auxiliary image,we can get the piexl which are fit for the hole region.Virtual view image generated by this method is better than others and the gap between the real point image and the target image become smaller.Relative to the bidrection DIBR technology,this approach does not require the depth information of auxiliary reference image, significantly reducing the amount of data. In addition, we do not process all the pixel of the virtual image, so the algorithm becomes simplifier and the computation time become shorter.